Content uploaded by Anton Beloglazov
Author content
All content in this area was uploaded by Anton Beloglazov
Content may be subject to copyright.
Energy Efficient Resource Management in Virtualized Cloud Data Centers
Anton Beloglazov* and Rajkumar Buyya
Cloud Computing and Distributed Systems (CLOUDS) Laboratory
Department of Computer Science and Software Engineering
The University of Melbourne, Australia
{abe, raj}@csse.unimelb.edu.au
Abstract—Rapid growth of the demand for computational
power by scientific, business and web-applications has led to
the creation of large-scale data centers consuming enormous
amounts of electrical power. We propose an energy efficient
resource management system for virtualized Cloud data centers
that reduces operational costs and provides required Quality
of Service (QoS). Energy savings are achieved by contin-
uous consolidation of VMs according to current utilization
of resources, virtual network topologies established between
VMs and thermal state of computing nodes. We present
first results of simulation-driven evaluation of heuristics for
dynamic reallocation of VMs using live migration according to
current requirements for CPU performance. The results show
that the proposed technique brings substantial energy savings,
while ensuring reliable QoS. This justifies further investigation
and development of the proposed resource management system.
Keywords-Energy efficiency; Cloud computing; Energy con-
sumption; Green IT; Resource management; Virtualization;
Allocation of virtual machines; Live migration of virtual
machines.
I. INTRODUCTION
Modern resource-intensive enterprise and scientific ap-
plications create growing demand for high performance
computing infrastructures. This has led to the construction
of large-scale computing data centers consuming enormous
amounts of electrical power. Despite of the improvements
in energy efficiency of the hardware, overall energy con-
sumption continues to grow due to increasing requirements
for computing resources. For example, in 2006 the cost
of energy consumption by IT infrastructures in US was
estimated as 4.5 billion dollars and it is likely to double
by 2011 [1]. Apart from the overwhelming operational
costs, building a data center leads to excessive establish-
ment expenses as data centers are usually built to serve
infrequent peak loads resulting in low average utilization of
the resources. Moreover, there are other crucial problems
that arise from high power consumption. Insufficient or
malfunctioning cooling system can lead to overheating of the
resources reducing system reliability and devices lifetime. In
addition, high power consumption by the infrastructure leads
* The author is in the first year of his candidature as a PhD student and
Prof. Rajkumar Buyya is the supervisor. The work presented in this paper
is a proposal for future research.
to substantial carbon dioxide (CO2) emissions contributing
to the greenhouse effect.
A number of practices can be applied to achieve energy-
efficiency, such as improvement of applications’ algorithms,
energy efficient hardware, Dynamic Voltage and Frequency
Scaling (DVFS) [2], terminal servers and thin clients, and
virtualization of computer resources [3]. Virtualization tech-
nology allows one to create several Virtual Machines (VMs)
on a physical server and, therefore, reduces amount of
hardware in use and improves the utilization of resources.
Among the benefits of virtualization are improved fault and
performance isolation between applications sharing the same
resource (a VM is viewed as a dedicated resource to the
customer); the ability to relatively easy move VMs from one
physical host to another using live or off-line migration; and
support for hardware and software heterogeneity.
Traditionally, an organization purchases its own comput-
ing resources and deals with maintenance and upgrade of
the outdated hardware, resulting in additional expenses. The
recently emerged Cloud computing paradigm [4] leverages
virtualization technology and provides the ability to pro-
vision resources on-demand on the pay-as-you-go basis.
Organizations can outsource their computation needs to the
Cloud, thereby eliminating the necessity to maintain own
computing infrastructure. Cloud computing naturally leads
to energy-efficiency by providing the following characteris-
tics:
•Economy of scale due to elimination of redundancies.
•Improved utilization of the resources.
•Location independence – VMs can be moved to a place
where energy is cheaper.
•Scaling up and down – resource usage can be adjusted
to current requirements.
•Efficient resource management by the Cloud provider.
One of the important requirements for a Cloud computing
environment is providing reliable QoS. It can be defined in
terms of Service Level Agreements (SLA) that describe such
characteristics as minimal throughput, maximal response
time or latency delivered by the deployed system. Although
modern virtualization technologies can ensure performance
isolation between VMs sharing the same physical computing
node, due to aggressive consolidation and variability of the
2010 10th IEEE/ACM International Conference on Cluster, Cloud and Grid Computing
978-0-7695-4039-9/10 $26.00 © 2010 IEEE
DOI 10.1109/CCGRID.2010.46
826
workload some VMs may not get the required amount of
resource when requested. This leads to performance loss
in terms of increased response time, time outs or failures
in the worst case. Therefore, Cloud providers have to deal
with energy-performance trade-off – minimization of energy
consumption, while meeting QoS requirements.
A. Research scope
The focus of this work is on energy-efficient resource
management strategies that can be applied on a virtualized
data center by a Cloud provider (e.g. Amazon EC2). The
main instrument that we leverage is live migration of VMs.
The ability to migrate VMs between physical hosts with
low overhead gives flexibility to a resource provider as VMs
can be dynamically reallocated according to current resource
requirements and the allocation policy. Idle physical nodes
can be switched off to minimize energy consumption.
In this paper we present a decentralized architecture of
the resource management system for Cloud data centers
and propose the development of the following policies for
continuous optimization of VM placement:
•Optimization over multiple system resources – at each
time frame VMs are reallocated according to current
CPU, RAM and network bandwidth utilization.
•Network optimization – optimization of virtual network
topologies created by intercommunicating VMs. Net-
work communication between VMs should be observed
and considered in reallocation decisions in order to
reduce data transfer overhead and network devices load.
•Thermal optimization – current temperature of physical
nodes is considered in reallocation decisions. The aim
is to avoid “hot spots” by reducing workload of the
overheated nodes and thus decrease error-proneness and
cooling system load.
B. Research challenges
The key challenges that have to be addressed are:
1) How to optimally solve the trade-off between energy
savings and delivered performance?
2) How to determine when, which VMs, and where to
migrate in order to minimize energy consumption by
the system, while minimizing migration overhead and
ensuring SLA?
3) How to develop efficient decentralized and scalable
algorithms for resource allocation?
4) How to develop comprehensive solution by combining
several allocation policies with different objectives?
The remainder of the paper is organized as follows. In
the next section we discuss related work followed by the
proposed system architecture in Section III. In Sections IV
and V we present the allocation policies and evaluation of
the heuristics followed by conclusion and future work.
II. RELATED WORK
Early work in energy aware resource management is
devoted to mobile devices with the objective of improving
battery lifetime [5], [6]. Later on, the context has been
shifted to data centers [7], [8] and virtual computing en-
vironments such as Clouds. Nathuji and Schwan [9] have
proposed an architecture of energy management system
for virtualized data centers where resource management
is divided into local and global policies. On the local
level, the system leverages guest operating system’s power
management strategies. Consolidation of VMs is handled by
global policies applying live migration to reallocate VMs.
However, the global policies are not discussed in detail
considering QoS requirements. In contrast, our work focuses
on global VM allocation policies considering strict SLA.
Kusic et al. [10] have stated the problem of continuous
consolidation as a sequential optimization and addressed
it using Limited Lookahead Control (LLC). The proposed
model requires simulation-based learning for the application-
specific adjustments. Due to complexity of the model the
optimization controller’s execution time reaches 30 minutes
even for a small number of nodes (e.g. 15), that is not
suitable for large-scale real-world systems. On the contrary,
our approach is heuristic-based allowing the achievement of
reasonable performance even for large-scale as shown in our
experimental studies.
Srikantaiah et al. [11] have studied the problem of re-
quests scheduling for multi-tiered web-applications in virtu-
alized heterogeneous systems in order to minimize energy
consumption, while meeting performance requirements. To
handle the optimization over multiple resources, the authors
have proposed a heuristic for multidimensional bin packing
problem as an algorithm for workload consolidation. How-
ever, the proposed approach is workload type and application
dependent, whereas our algorithms are independent of the
workload type and thus are suitable for a generic Cloud
environment.
Song et al. [12] have proposed resource allocation to
applications according to their priorities in multi-application
virtualized cluster. The approach requires machine-learning
to obtain utility functions for the applications and defined
application priorities. Unlike our work, it does not apply
migration of VMs to optimize allocation continuously (the
allocation is static).
Cardosa et al. [13] have explored the problem of power-
efficient allocation of VMs in virtualized heterogeneous
computing environments. They have leveraged “min”, “max”
and “shares” parameters of VMM that represent minimum,
maximum and proportion of CPU allocated to VMs sharing
the same resource. The approach suits only enterprise envi-
ronments or private Clouds as it does not support strict SLA
and requires knowledge of applications priorities to define
shares parameter.
827
Verma et al. [14] have formulated the problem of dy-
namic placement of applications in virtualized heteroge-
neous systems as continuous optimization: at each time
frame the placement of VMs is optimized to minimize power
consumption and maximize performance. The authors have
applied a heuristic for bin packing problem with variable
bin sizes and costs. The authors have introduced the notion
of cost of VM live migration, but the information about the
cost calculation is not provided. The proposed algorithms,
on the contrary to our approach, do not handle strict SLA
requirements: SLA can be violated due to variability of the
workload.
In contrast to the discussed studies, we propose efficient
heuristics for dynamic adaption of allocation of VMs in run-
time applying live migration according to current utilization
of resources and thus minimizing energy consumption. The
proposed approach can effectively handle strict QoS re-
quirements, heterogeneous infrastructure and heterogeneous
VMs. The algorithms do not depend on a particular type of
workload and do not require any knowledge about applica-
tions executing on VMs.
Calheiros et al. [15] have investigated the problem of
mapping VMs on physical nodes optimizing network com-
munication between VMs, however, the problem has not
been explored in the context of energy consumption min-
imization. Recently, a number of research works have been
done on thermal efficient resource management in data
centers [16], [17]. The studies show that software-driven
thermal management and temperature aware workload place-
ment bring additional energy savings. However, the problem
of thermal management in the context of virtualized data
centers has not been investigated. Moreover, to the best of
out knowledge there are no studies on a comprehensive
approach that combines optimization of VM placement
according to current utilization of the resources with net-
work and thermal optimization for virtualized data centers.
Therefore, the exploration of such approach is timely and
crucial, especially considering rapid development of Cloud
computing environments.
III. SYSTEM ARCHITECTURE
In this work the underlying infrastructure is represented by
a large-scale Cloud data center comprising nheterogeneous
physical nodes. Each node has a CPU, which can be multi-
core, with performance defined in Millions Instructions Per
Second (MIPS). Besides that, a node is characterized by
the amount of RAM and network bandwidth. Users submit
requests for provisioning of mheterogeneous VMs with
resource requirements defined in MIPS, amount of RAM
and network bandwidth. SLA violation occurs when a VM
cannot get the requested amount of resource, which may
happen due to VM consolidation.
The software system architecture is tiered comprising a
dispatcher, global and local managers. The local managers
reside on each physical node as a part of a Virtual Machine
Monitor (VMM). They are responsible for observing current
utilization of the node’s resources and its thermal state. The
local managers choose VMs that have to be migrated to
another node in the following cases:
•The utilization of some resource is close to 100% that
creates a risk of SLA violation.
•The utilization of resources is low, therefore, all the
VMs should be reallocated to another node and the idle
node should be turned off.
•A VM has intensive network communication with an-
other VM allocated to a different physical host.
•The temperature exceeds some limit and VMs have to
be migrated in order to reduce load on the cooling
system and allow the node to cool down naturally.
The local managers send to the global managers the
information about the utilization of resources and VMs
chosen to migrate. Besides that, they issue commands for
VM resizing, application of DVFS and turning on / off idle
nodes. Each global manager is attached to a set of nodes
and processes data obtained from their local managers. The
global managers continuously apply distributed version of
a heuristic for semi-online multidimensional bin-packing,
where bins represent physical nodes and items are VMs
that have to be allocated. The decentralization removes
a Single Point of Failure (SPF) and improves scalability.
Each dimension of an item represents the utilization of a
particular resource. After obtaining allocation decision, the
global managers issue commands for live migration of VMs.
As shown in Figure 1, the system operation consists of
the following steps:
1) New requests for VM provisioning. Users submit
requests for provisioning of VMs.
2) Dispatching requests for VM provisioning. The dis-
patcher distributes requests among global managers.
3) Intercommunication between global managers. The
global managers exchange information about utiliza-
tion of resources and VMs that have to be allocated.
4) Data about utilization of resources and VMs chosen
to migrate. The local managers propagate information
about resource utilization and VMs chosen to migrate
to the global managers.
5) Migration commands. The global managers issue
VM migration commands in order to optimize current
allocation.
6) Commands for VM resizing and adjusting of power
states. The local managers monitor their host nodes
and issue commands for VM resizing and changes in
power states of nodes.
7) VM resizing, scheduling and migration actions.
According to the received commands, VMM performs
actual resizing and migration of VMs as well as
resource scheduling.
828
Physical node n
VM VM VM
Local Manager VMM
7
7
7
6
Dispatcher
4
2
...
7
1
Users 2
Global Manager
5
Physical node 1
VM VM VM
Local Manager VMM
7
7
7
6
4
...
7
Global Manager
5
Physical node 2
VM VM VM
Local Manager VMM
7
7
7
6
...
7
...
5 54 4
3
...
Figure 1. The system architecture
IV. ALLOCATION POLICIES
We propose three stages of VM placement optimization:
reallocation according to current utilization of multiple sys-
tem resources, optimization of virtual network topologies
established between VMs and VM reallocation considering
thermal state of the resources. Each of these stages is
planned to be investigated separately and then combined in
an overall solution. The developed algorithms have to meet
the following requirements:
•Decentralization and parallelism – to eliminate SPF and
provide scalability.
•High performance – the system has to be able to quickly
respond to changes in the workload.
•Guaranteed QoS – the algorithms have to provide
reliable QoS by meeting SLA.
•Independence of the workload type – the algorithms
have to be able to perform efficiently in mixed-
application environments.
The VM reallocation problem can be divided in two:
selection of VMs to migrate and determining new placement
of these VMs on physical hosts. The first part has to
be considered separately for each optimization stage. The
second part is solved by application of a heuristic for semi-
online multidimensional bin-packing problem.
At the first optimization stage, the utilization of resources
is monitored and VMs are reallocated to minimize the num-
ber of physical nodes in use and thus minimize energy con-
sumption by the system. However, aggressive consolidation
of VMs may lead to violation of performance requirements.
We have proposed several heuristics for selection of VMs to
migrate and investigated the trade-off between performance
and energy savings. To simplify the problem for the first
step we considered only utilization of CPU. The main idea
of the policies is to set upper and lower utilization thresholds
and keep total utilization of CPU created by VMs sharing
the same node between these thresholds. If the utilization
exceeds the upper thresholds, some VMs have to be migrated
from the node to reduce the risk of SLA violation. If the
utilization goes bellow the lower thresholds, all VMs have
to be migrated and the node has to be switched off to save
the energy consumed be the idle node. Another problem is to
determine particular values of the utilization thresholds. The
results of the proposed algorithms evaluation are presented
in Section V.
Due to continuous reallocation, some intensively com-
municating VMs can be placed inefficiently leading to
excessive load on the network facilities. Therefore, it is
crucial to consider network communication behavior of VMs
in reallocation decisions. The aim of the second proposed
optimization stage is to place communicating VMs in a way
minimizing the overhead of data transfer over network.
A cooling system of a data center consumes a significant
amount of energy, therefore, the third proposed optimization
stage is aimed at optimization of cooling system operation.
Due to consolidation, some computing nodes experience
high load leading to overheating and thus require extensive
cooling. Monitoring of the nodes’ thermal state using sensors
gives an opportunity to recognize overheating and reallocate
workload from the overheated node to allow the natural
cooling. The network and temperature optimizations are
subjects for the ongoing research work.
V. E VALUATION
As the proposed system is targeted on a large-scale Cloud
data center, it is necessary to conduct large-scale experi-
ments to evaluate the algorithms. However, it is difficult to
run large-scale experiments on a real-world infrastructure,
especially when the experiments have to be repeated for
different policies with the same conditions [18]. Therefore,
simulation has been chosen as a way to evaluate the pro-
posed heuristics. We have chosen CloudSim toolkit [18] as
a simulation framework, as it is built for simulation of Cloud
computing environments. In comparison to alternative simu-
lation toolkits (e.g. SimGrid, GangSim), CloudSim supports
modeling of on-demand virtualization enabled resource and
829
application management. We have extended the framework
in order to enable energy aware simulations as the core
framework does not provide this capability. In addition, we
have incorporated the abilities to account SLA violations
and to simulate dynamic workloads that correspond to web-
applications and online services.
The simulated data center consists of 100 heterogeneous
physical nodes. Each node is modeled to have one CPU core
with performance equivalent to 1000, 2000 or 3000 MIPS,
8 Gb of RAM and 1 TB of storage. Users submit requests
for provisioning of 290 heterogeneous VMs that fill the full
capacity of the data center. For the borderline policies we
simulated a Non Power Aware policy (NPA) and DVFS
that adjusts the voltage and frequency of CPU according
to current utilization. We simulated a Single Threshold
policy (ST) and two-threshold policy aimed at Minimization
of Migrations (MM). Besides that, the policies have been
evaluated with different values of the thresholds.
Table I
SIMULATION RESULTS
Policy Energy SLA Migr. Avg. SLA
NPA 9.15 KWh - - -
DVFS 4.40 KWh - - -
ST 50% 2.03 KWh 5.41% 35 226 81%
ST 60% 1.50 KWh 9.04% 34 231 89%
MM 30-70% 1.48 KWh 1.11% 3 359 56%
MM 40-80% 1.27 KWh 2.75% 3 241 65%
MM 50-90% 1.14 KWh 6.69% 3 120 76%
The simulation results are presented in Table I. The
results show that dynamic reallocation of VMs according to
current utilization of CPU can bring higher energy savings
comparing to static allocation policies. MM policy allows
to achieve the best energy savings: by 83%, 66% and
23% less energy consumption relatively to NPA, DVFS
and ST policies respectively with thresholds 30-70% and
ensuring percentage of SLA violations of 1.1%; and by 87%,
74% and 43% with thresholds 50-90% and 6.7% of SLA
violations. MM policy leads to more than 10 times fewer
VM migrations than ST. The results show the flexibility of
the algorithm, as the thresholds can be adjusted according to
SLA requirements. Strict SLA (1.11%) allow achievement of
the energy consumption of 1.48 KWh. However, if SLA are
relaxed (6.69%), the energy consumption is further reduced
to 1.14 KWh.
VI. CONCLUSION AND FUTURE WORK
In this paper have presented a decentralized architecture
of the energy aware resource management system for Cloud
data centers. We have defined the problem of minimizing
the energy consumption while meeting QoS requirements
and stated the requirements for VM allocation policies.
Moreover, we have proposed three stages of continuous
optimization of VM placement and presented heuristics for
a simplified version of the first stage. The heuristics have
been evaluated by simulation using the extended CloudSim
toolkit. One of the heuristics leads to significant reduction
of the energy consumption by a Cloud data center – by
83% in comparison to a non power aware system and by
66% in comparison to a system that applies only DVFS
technique but does not adapt allocation of VMs in run-time.
Moreover, MM policy enables flexible adjustment of SLA by
setting appropriate values of the utilization thresholds: SLA
can be relaxed leading to further improvement of energy
consumption. The policy supports heterogeneity of both the
hardware and VMs and does not require any knowledge
about particular applications running on the VMs. The policy
is independent of the workload type.
Table II
FUTURE RESEARCH WORK AND TIMELINE
Timeline Work description
01/2010 - 03/2010 Completion of the algorithms for the opti-
mization over multiple resources.
04/2010 - 09/2010 Development of the algorithms and a proto-
type for the network optimization.
10/2010 - 03/2011 Development of the algorithms and a proto-
type for the temperature optimization.
04/2011 - 02/2012 Overall solution and a real implementation
of the system as a module of a VMM, and
experimental evaluation.
We propose a plan for the future research work that
consists of several steps presented in Table II. Once the
algorithms for all of the proposed optimization stages are
developed, they will be combined in an overall solution and
implemented as a part of a real-world Cloud platform, such
as Aneka1.
The obtained results show that dynamic consolidation
of VMs brings substantial energy savings while providing
required QoS. Besides the significant reduction of opera-
tional costs, the project is socially valuable as it decreases
carbon dioxide footprints and overall energy consumption
by modern IT infrastructures.
REFERENCES
[1] R. Brown et al., “Report to congress on server and data center
energy efficiency: Public law 109-431,” Lawrence Berkeley
National Laboratory, 2008.
[2] G. Semeraro, G. Magklis, R. Balasubramonian, D. H. Al-
bonesi, S. Dwarkadas, and M. L. Scott, “Energy-efficient
processor design using multiple clock domains with dynamic
voltage and frequency scaling,” in Proceedings of the 8th
International Symposium on High-Performance Computer
Architecture, 2002, pp. 29–42.
1Aneka is a market-oriented Cloud development and management plat-
form with rapid application development and workload distribution capa-
bilities (http://www.manjrasoft.com/).
830
[3] P. Barham, B. Dragovic, K. Fraser, S. Hand, T. Harris, A. Ho,
R. Neugebauer, I. Pratt, and A. Warfield, “Xen and the art of
virtualization,” in Proceedings of the 19th ACM symposium
on Operating systems principles, 2003, p. 177.
[4] R. Buyya, C. S. Yeo, and S. Venugopal, “Market-oriented
cloud computing: Vision, hype, and reality for delivering
it services as computing utilities,” in Proceedings of the
10th IEEE International Conference on High Performance
Computing and Communications (HPCC’08). IEEE CS
Press, Los Alamitos, CA, USA, 2008.
[5] R. Neugebauer and D. McAuley, “Energy is just another re-
source: Energy accounting and energy pricing in the nemesis
OS,” in Proceedings of the 8th IEEE Workshop on Hot Topics
in Operating Systems, 2001, pp. 59–64.
[6] H. Zeng, C. S. Ellis, A. R. Lebeck, and A. Vahdat, “ECOSys-
tem: managing energy as a first class operating system
resource,” ACM SIGPLAN Notices, vol. 37, no. 10, p. 132,
2002.
[7] E. Pinheiro, R. Bianchini, E. V. Carrera, and T. Heath,
“Load balancing and unbalancing for power and performance
in cluster-based systems,” in Workshop on Compilers and
Operating Systems for Low Power, 2001, pp. 182–195.
[8] J. S. Chase, D. C. Anderson, P. N. Thakar, A. M. Vahdat,
and R. P. Doyle, “Managing energy and server resources in
hosting centers,” in Proceedings of the 18th ACM symposium
on Operating systems principles. ACM New York, NY, USA,
2001, pp. 103–116.
[9] R. Nathuji and K. Schwan, “Virtualpower: Coordinated power
management in virtualized enterprise systems,” ACM SIGOPS
Operating Systems Review, vol. 41, no. 6, pp. 265–278, 2007.
[10] D. Kusic, J. O. Kephart, J. E. Hanson, N. Kandasamy, and
G. Jiang, “Power and performance management of virtual-
ized computing environments via lookahead control,” Cluster
Computing, vol. 12, no. 1, pp. 1–15, 2009.
[11] S. Srikantaiah, A. Kansal, and F. Zhao, “Energy aware con-
solidation for cloud computing,” Cluster Computing, vol. 12,
pp. 1–15, 2009.
[12] Y. Song, H. Wang, Y. Li, B. Feng, and Y. Sun, “Multi-Tiered
On-Demand resource scheduling for VM-Based data center,”
in Proceedings of the 2009 9th IEEE/ACM International
Symposium on Cluster Computing and the Grid-Volume 00,
2009, pp. 148–155.
[13] M. Cardosa, M. Korupolu, and A. Singh, “Shares and util-
ities based power consolidation in virtualized server envi-
ronments,” in Proceedings of IFIP/IEEE Integrated Network
Management (IM), 2009.
[14] A. Verma, P. Ahuja, and A. Neogi, “pMapper: power and
migration cost aware application placement in virtualized
systems,” in Proceedings of the 9th ACM/IFIP/USENIX Inter-
national Conference on Middleware. Springer-Verlag New
York, Inc., 2008, pp. 243–264.
[15] R. N. Calheiros, R. Buyya, and C. A. F. D. Rose, “A
heuristic for mapping virtual machines and links in emulation
testbeds,” 2009.
[16] R. K. Sharma, C. E. Bash, C. D. Patel, R. J. Friedrich, and
J. S. Chase, “Balance of power: Dynamic thermal manage-
ment for internet data centers,” IEEE Internet Computing, pp.
42–49, 2005.
[17] J. Moore, J. Chase, P. Ranganathan, and R. Sharma, “Making
scheduling” cool”: temperature-aware workload placement in
data centers,” 2005.
[18] R. Buyya, R. Ranjan, and R. N. Calheiros, “Modeling and
simulation of scalable cloud computing environments and the
CloudSim toolkit: Challenges and opportunities,” in Proceed-
ings of the 7th High Performance Computing and Simulation
Conference (HPCS’09). IEEE Press, NY, USA, 2009.
831
